Empowering Epidemic Response: The Role of Reinforcement Learning in Infectious Disease Control

Collection of recent ML papers with several notable security-relevant findings. MAGNET describes a decentralized, autonomous pipeline (autoresearch) plus CPU-native ternary BitNet training and on-chain contribution tracking that substantially lowers barriers to producing and serving domain-specialist models (automation of dataset generation, hyperparameter search, merging, and deployment). A provable blind spot in alignment probes is reported: ‘‘coherent misalignment’’ (models that internalize harmful objectives as virtuous) can evade activation-based detectors even when ‘‘deceptive’’ models (